Gal Radia

Paste a link. Get a cookable recipe.

Turns a recipe buried in an Instagram post, a Facebook post, or any recipe site into a clean, structured, cookable format. A layered pipeline reads structured page data first, falls back to a printer-friendly parse, then hands the hard cases to an LLM. One codebase ships as the website and a native Android app. Hebrew-first and RTL by default, with an always-dark Cook Mode for use at the stove.

• ▸Scrapes Instagram, Facebook, and general recipe sites, tailored per platform
• ▸Falls back through JSON-LD, printer page, then LLM, balancing cost and accuracy
• ▸Swaps AI provider freely: Gemini or self-hosted Ollama
• ▸Ships web and native Android from one codebase via Capacitor
• ▸Runs Hebrew-first on an RTL-native design system
• ▸Includes an always-dark Cook Mode with built-in timers

React 19 · TypeScript · Vite · Tailwind CSS v4 · Framer Motion · TanStack Query · Zustand · Capacitor · ASP.NET Core · EF Core · PostgreSQL · Supabase · Google Gemini · Ollama · PuppeteerSharp · Turborepo

Snap a receipt. See where your money goes.

Turns a photo of a paper receipt into an editable record: shop, date, line items, discounts, total. Gemini 2.5 Flash-Lite and Cloud Vision OCR read it in about 4 seconds. Nothing saves automatically, every read lands in a review step first. Past that it tracks spending by category, month, shop, and day of week. Fully bilingual Hebrew and English, tuned for Israeli receipts.

• ▸Reads receipts with Gemini 2.5 Flash-Lite plus Cloud Vision OCR in ~4s
• ▸Routes every AI read through an editable review step before saving
• ▸Searches down to the line item across every scanned receipt
• ▸Charts spending by category, month, shop, and day of week
• ▸Runs fully bilingual in Hebrew and English, RTL-correct
• ▸Hands data back on demand: CSV export and full account deletion

React 19 · TypeScript · Vite · Tailwind CSS v4 · shadcn/ui · Radix UI · Recharts · i18next · Capacitor · Firebase · Python · FastAPI · Google Gemini 2.5 · Cloud Vision OCR · Cloud Run

Machine learning that reprioritizes Linux processes in real time.

Watches processes as they start and exit, predicts how long each one runs, and rewrites its nice value before it competes for CPU. eBPF programs on the kernel tracepoints capture events at low overhead. Each binary gets fingerprinted by ELF section sizes, memory, and I/O, then XGBoost predicts turnaround time and K-means assigns a priority class. Every decision logs to SQLite for retraining. Cuts average turnaround time by over 5%.

• ▸Monitors process exec and exit through eBPF tracepoints at low overhead
• ▸Predicts turnaround time with XGBoost before a process competes for CPU
• ▸Assigns a priority class with K-means, mapped to nice ranges
• ▸Fingerprints binaries by ELF section sizes, memory, and I/O
• ▸Logs every prediction and outcome to SQLite for retraining
• ▸Cuts average turnaround time by over 5% against baseline

Python · C · eBPF / BCC · XGBoost · scikit-learn · K-means · pandas · NumPy · SQLite · psutil · Linux kernel